The current study addresses the question of how word-level (“stress”) and phrase- or sentence-level prominence (“accent”) is realized in Polish. For this purpose, a production experiment eliciting semi-spontaneous utterances was conducted, closely following the methodological approach introduced in [1]. Our acoustic analyses are based on identical target syllables which are embedded in sentences under conditions that allow to disentangle word-level and phrase-level prominence. The acoustic realizations of these target syllables are then subject to linear mixed-effect models fitted for various acoustic parameters: duration, fundamental frequency maximum, intensity, and spectral balance. The models indicate that prominence marking in Polish is realized acoustically in a stable fashion on phrase-level only. Word stress marking occurs only in cases where a lexically stressed syllable simultaneously realizes a phrase-level accent.